[LUCENE-948] Writers on two machines over NFS can hit FNFE due to stale NFS client caching

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.1
• Fix Version/s: 2.3
• Component/s: core/index
• Lucene Fields: New
• Assignee: Michael McCandless
• Reporter: Michael McCandless
• Created: 03/Jul/07 11:29
• Updated: 28/Aug/22 11:39
• Resolved: 03/Aug/07 18:49

Description

Issue spawned from this thread:

http://www.gossamer-threads.com/lists/lucene/java-user/50680

When IndexFileDeleter lists the directory, looking for segments_X
files to load, if it hits a FNFE on opening such a file it should
catch this and treat it as if the file does not exist.

On NFS (and possibly other file systems), a directory listing is not
guaranteed to be "current"/coherent. Specifically, if machine #1 has
just removed file "segments_n" and shortly thereafer machine #2 does a
dir listing, it's possible (likely?) that the dir listing will still
show that segments_n exists.

I think the fix is simple: catch the FNFE and just handle it as if the
segments_n does not in fact exist.